An overview of the analytical approach underpinning the HUMANITY project, including key interpretive considerations for understanding the findings.
Temporal patterns should be interpreted with caution. The composition of documents in the corpus varies across years, reflecting both the evolution of each crisis and shifts in humanitarian reporting practices. Although the models account for document type, agency, and year as prevalence covariates, temporal trends cannot be fully disentangled from changes in the types of documents produced during different crisis phases. Year-based patterns are therefore best understood as capturing the discursive evolution of humanitarian governance within changing operational contexts, rather than isolated shifts in behaviour. Variation in sentence coverage across years also affects the precision of temporal estimates.
Expected topic proportions reflect the relative discursive salience of themes within the corpus. They should not be interpreted as direct measures of issue frequency, policy priority, or substantive importance in the field. These proportions indicate how prominently a given theme features in the documented record, which is shaped by reporting conventions, donor requirements, and organisational mandates as much as by operational realities on the ground.
The findings are derived from publicly available humanitarian documents and are therefore bounded by what organisations choose to report and publish. Gaps in the corpus — whether due to access constraints, security concerns, or institutional reporting norms — should be considered when interpreting the results. The analysis captures documented discourse rather than the full scope of humanitarian practice.
A detailed description of the data collection, structural topic modelling approach, and analytical framework will be published here as the project progresses.
Read About the Project →